Skip to content
Part IA Michaelmas, Lent Term

Diagonalisation and Cantor's Theorem

Cantor’s Theorem

Statement

Theorem (Cantor, 1891): For any set AA, there is no surjection AP(A)A \twoheadrightarrow \mathcal{P}(A).

Corollary

P(A)>A|\mathcal{P}(A)| > |A| for every set AA.

In particular, there are different sizes of infinity.

Examinable Proof (Diagonalisation)

Setup

Suppose e:AP(A)e : A \twoheadrightarrow \mathcal{P}(A) is surjective.

The Diagonal Set

Define UAU \subseteq A by:

U={xA:xe(x)}U = \{x \in A : x \notin e(x)\}

This is the set of all elements that are not members of the set they’re mapped to.

Contradiction

Since ee is surjective, there exists aAa \in A with e(a)=Ue(a) = U.

Question: Is aUa \in U?

  • If aUa \in U: Then ae(a)=Ua \notin e(a) = U. Contradiction.
  • If aUa \notin U: Then ae(a)=Ua \in e(a) = U (by definition of UU). Contradiction.

Either way, contradiction. Therefore no such surjection exists.

Visualisation

Consider A=NA = \mathbb{N}. If e:NP(N)e : \mathbb{N} \to \mathcal{P}(\mathbb{N}) enumerated all subsets:

nne(n)Ne(n) \subseteq \mathbb{N}ne(n)n \in e(n)?
0{1,2,3,}\{1, 2, 3, \ldots\}No
1{0,2,4,}\{0, 2, 4, \ldots\}No
2{0,1,3,}\{0, 1, 3, \ldots\}Yes
3{3}\{3\}Yes
\vdots\vdots\vdots

The diagonal UU “flips” the membership on the diagonal. UU differs from every row, so cannot appear.

Cantor diagonalisation visualisation

Immediate Consequences

Uncountability of R\mathbb{R}

RP(N)\mathbb{R} \cong \mathcal{P}(\mathbb{N}) via binary expansions.

Since P(N)\mathcal{P}(\mathbb{N}) is uncountable, R\mathbb{R} is uncountable.

No Largest Cardinality

Given any set AA, the set P(A)\mathcal{P}(A) is strictly larger.

Iterating: A<P(A)<P(P(A))<|A| < |\mathcal{P}(A)| < |\mathcal{P}(\mathcal{P}(A))| < \ldots

NR\mathbb{N} \subsetneq \mathbb{R}

N\mathbb{N} is countable, R\mathbb{R} is not.

Lawvere’s Fixed Point Theorem

Statement

If there exists a surjection e:A(AX)e : A \twoheadrightarrow (A \to X), then every function f:XXf : X \to X has a fixed point.

Proof

Define φ:AX\varphi : A \to X by φ(a)=f(e(a)(a))\varphi(a) = f(e(a)(a)).

Since ee is surjective, there exists a0a_0 with e(a0)=φe(a_0) = \varphi.

Then: e(a0)(a0)=φ(a0)=f(e(a0)(a0))e(a_0)(a_0) = \varphi(a_0) = f(e(a_0)(a_0)).

So e(a0)(a0)e(a_0)(a_0) is a fixed point of ff.

Deriving Cantor’s Theorem

Take X={0,1}X = \{0, 1\}.

A surjection A(A{0,1})A \twoheadrightarrow (A \to \{0, 1\}) would imply every f:{0,1}{0,1}f : \{0, 1\} \to \{0, 1\} has a fixed point.

But the swap map f(0)=1f(0) = 1, f(1)=0f(1) = 0 has no fixed point. Contradiction.

Therefore (A{0,1})P(A)(A \to \{0, 1\}) \cong \mathcal{P}(A) has no surjection from AA.


Summary

  • Cantor’s Theorem: No surjection AP(A)A \twoheadrightarrow \mathcal{P}(A)
  • Proof uses diagonal argument: construct U={x:xe(x)}U = \{x : x \notin e(x)\}
  • R\mathbb{R} is uncountable; there is no largest set
  • Lawvere’s theorem generalises Cantor: surjections to function spaces force fixed points